Risk prediction for chronic atrophic gastritis using a random forest model: A multicenter study

Figure 1 Study flowchart. A total of 1375 subjects were initially recruited from a gastric cancer screening program. After participants were excluded because of missing clinical data or a diagnosis of gastric cancer, 1268 subjects were included in the final analysis. The dataset was split into a training set (75%) and a testing set (25%). Feature selection was performed using least absolute shrinkage and selection operator regression, followed by the construction of a random forest model using 5-fold cross-validation. The model's performance was evaluated, and its predictions were interpreted using SHapley Additive exPlanation. Finally, the model was validated on an independent external cohort of 120 subjects. GC: Gastric cancer; LASSO: Least absolute shrinkage and selection operator; CV: Cross-validation; AUC: Area under the curve; SHAP: SHapley Additive exPlanation.